Grammar of Graphics and ggplot2
SISBID 2016
Di Cook (dicook@monash.edu, @visnut); Heike Hofmann (heike.hofmann@gmail.com, @heike_hh)i; edited by Alexej Gossmann
2016-07-14
Using the package ggplot2
Elements of a plot
- data
- aesthetics: mapping of variables to graphical elements
- geom: type of plot structure to use
- transformations: log scale, …
Additional components
- layers: multiple geoms, multiple data sets, annotation
- facets: show subsets in different plots
- themes: modifying style
Why use a grammar of graphics?
Variable in the data is directly mapped to an element in the plot
Data - Autism
glimpse(autism)
# Observations: 604
# Variables: 7
# $ childid <int> 1, 1, 1, 1, 1, 10, 10, 10, 10, 100, 100, 100, 100, 10...
# $ sicdegp <fctr> high, high, high, high, high, low, low, low, low, hi...
# $ age2 <dbl> 0, 1, 3, 7, 11, 0, 1, 7, 11, 0, 1, 3, 7, 0, 1, 7, 11,...
# $ vsae <int> 6, 7, 18, 25, 27, 9, 11, 18, 39, 15, 24, 37, 135, 8, ...
# $ gender <fctr> male, male, male, male, male, male, male, male, male...
# $ race <fctr> white, white, white, white, white, white, white, whi...
# $ bestest2 <fctr> pdd, pdd, pdd, pdd, pdd, autism, autism, autism, aut...Plotting points
ggplot(autism, aes(x=age2, y=vsae)) +
geom_point()
Your turn
How is the data mapped to graphical elements?
- data: autism
- aesthetics: x and y positions along a line
- geom: point
- transformations: none
Jittering points
ggplot(autism, aes(x=age2, y=vsae)) +
geom_jitter()
Your turn
How is the data mapped to graphical elements?
- data: autism
- aesthetics: x and y positions along a line
- geom: point
- transformations: jitter
Adding lines
ggplot(autism, aes(x=age2, y=vsae)) +
geom_point() + geom_line()
Not the lines we want
ggplot(autism, aes(x=age2, y=vsae, group=childid)) +
geom_point() + geom_line()
Too much ink
ggplot(autism, aes(x=age2, y=vsae, group=childid)) +
geom_point() + geom_line(alpha=0.5)
ggplot(autism, aes(x=age2, y=vsae, group=childid)) +
geom_line(alpha=0.2) + theme_bw()
Now we can see that some individuals degrade, while most improve with time.
Log scale y?
ggplot(autism, aes(x=age2, y=vsae, group=childid)) +
geom_line(alpha=0.2) + scale_y_log10() + theme_bw()
By age 2 diagnosis
ggplot(autism, aes(x=age2, y=vsae, group=childid, colour=bestest2)) +
geom_line(alpha=0.3) + scale_y_log10() + theme_bw()
Now we can see a lot of overlap between the two groups.
Refine groups
ggplot(autism, aes(x=age2, y=vsae, colour=bestest2)) +
geom_point(alpha=0.1) + geom_line(aes(group=childid), alpha=0.1) +
geom_smooth(se=F) + scale_y_log10() + theme_bw()
ggplot(autism, aes(x=age2, y=vsae, colour=bestest2)) +
geom_point(alpha=0.1) + geom_line(aes(group=childid), alpha=0.1) +
geom_smooth(se=F, method=lm) + scale_y_log10() + theme_bw()
Your turn
What do we learn about autism, age, and the diagnosis at age 2?
In terms of categorisation into either pdd or autism the vsae score is not distinct, but on average the autism diagnosis 2 year olds have lower scores. There a lot of overlap between ths group.
Your turn
How is the data mapped to graphical elements?
- data: autism
- aesthetics: age on x and vsae on y, bestest2 mapped to color, smoothed vsae mapped to y
- geom: point, line
- transformations: log10 on y
A different look
ggplot(autism, aes(x=age2, y=vsae, colour=bestest2)) +
geom_boxplot() + scale_y_log10()
That’s not what I wanted ….
For each age measured
ggplot(autism, aes(x=factor(age2), y=vsae, colour=bestest2)) +
geom_boxplot() + scale_y_log10()
Which is better?
p1 <- ggplot(autism, aes(x=age2, y=vsae, colour=bestest2)) +
geom_point(alpha=0.1) + geom_line(aes(group=childid), alpha=0.1) +
geom_smooth(se=F) + scale_y_log10() + theme(legend.position="none")
p2 <- ggplot(autism, aes(x=factor(age2), y=vsae, colour=bestest2)) +
geom_boxplot() + scale_y_log10() + theme(legend.position="none")
grid.arrange(p1, p2, ncol=2)
New example - Flying etiquette
41% Of Fliers Think You’re Rude If You Recline Your Seat
fly <- read_csv("./data/flying-etiquette.csv")
glimpse(fly)
# Observations: 1,040
# Variables: 27
# $ RespondentID <dbl> ...
# $ How often do you travel by plane? <chr> ...
# $ Do you ever recline your seat when you fly? <chr> ...
# $ How tall are you? <int> ...
# $ Do you have any children under 18? <chr> ...
# $ In a row of three seats, who should get to use the two arm rests? <chr> ...
# $ In a row of two seats, who should get to use the middle arm rest? <chr> ...
# $ Who should have control over the window shade? <chr> ...
# $ Is itrude to move to an unsold seat on a plane? <chr> ...
# $ Generally speaking, is it rude to say more than a few words tothe stranger sitting next to you on a plane? <chr> ...
# $ On a 6 hour flight from NYC to LA, how many times is it acceptable to get up if you're not in an aisle seat? <chr> ...
# $ Under normal circumstances, does a person who reclines their seat during a flight have any obligation to the person sitting behind them? <chr> ...
# $ Is itrude to recline your seat on a plane? <chr> ...
# $ Given the opportunity, would you eliminate the possibility of reclining seats on planes entirely? <chr> ...
# $ Is it rude to ask someone to switch seats with you in order to be closer to friends? <chr> ...
# $ Is itrude to ask someone to switch seats with you in order to be closer to family? <chr> ...
# $ Is it rude to wake a passenger up if you are trying to go to the bathroom? <chr> ...
# $ Is itrude to wake a passenger up if you are trying to walk around? <chr> ...
# $ In general, is itrude to bring a baby on a plane? <chr> ...
# $ In general, is it rude to knowingly bring unruly children on a plane? <chr> ...
# $ Have you ever used personal electronics during take off or landing in violation of a flight attendant's direction? <chr> ...
# $ Have you ever smoked a cigarette in an airplane bathroom when it was against the rules? <chr> ...
# $ Gender <chr> ...
# $ Age <chr> ...
# $ Household Income <chr> ...
# $ Education <chr> ...
# $ Location (Census Region) <chr> ...Variables
Mix of categorical and quantiative variables. What mappings are appropriate? Area for counts of categories, side-by-side boxplots for mixed pair.
Support
ggplot(fly, aes(x=`How often do you travel by plane?`)) +
geom_bar() + coord_flip()
Categories are not sorted
Sorted categories
fly$`How often do you travel by plane?` <-
factor(fly$`How often do you travel by plane?`, levels=c(
"Never","Once a year or less","Once a month or less",
"A few times per month","A few times per week","Every day"))
ggplot(fly, aes(x=`How often do you travel by plane?`)) +
geom_bar() + coord_flip()
Filter data
fly_sub <- fly %>% filter(`How often do you travel by plane?` %in%
c("Once a year or less",
"Once a month or less")) %>%
filter(!is.na(`Do you ever recline your seat when you fly?`)) %>%
filter(!is.na(Age)) %>% filter(!is.na(Gender))Recline by height
fly_sub$`Do you ever recline your seat when you fly?` %>% unique()
# [1] "About half the time" "Usually" "Always"
# [4] "Once in a while" "Never"
fly_sub$`Do you ever recline your seat when you fly?` <- factor(
fly_sub$`Do you ever recline your seat when you fly?`, levels=c(
"Never","Once in a while","About half the time",
"Usually","Always"))
ggplot(fly_sub, aes(y=`How tall are you?`,
x=`Do you ever recline your seat when you fly?`)) +
geom_boxplot() #+ coord_flip()
Cheat sheet
Take a look at the ggplot2 Cheat sheet
Your turn
How many geoms are available in ggplot2? What is geom_rug?
p <- ggplot(autism, aes(x=age2, y=vsae))
p1 <- p + geom_point() + coord_flip()
p2 <- p + geom_point() + geom_rug() + coord_flip()
p3 <- p + geom_point() + geom_rug(position='jitter') + coord_flip()
grid.arrange(p1, p2, p3, nrow=3)
Your turn
What is the difference between colour and fill?
Colour is for 0 or 1-dimensional elements, and fill is for area (2-d) geoms
Your turn
What does coord_fixed() do? What is the difference between this and using theme(aspect.ratio=...)?
p <- ggplot(autism, aes(x=age2, y=vsae))
p1 <- p + geom_point() + coord_fixed(ratio = 1)
p2 <- p + geom_point() + theme(aspect.ratio = 1)
grid.arrange(p1, p2, ncol=2)
coord_fixed operates on the raw data values, but theme(aspect_ratio=...) works on the plot dimensions.
Your turn
What are scales? How many numeric transformation scales are there?
scales do the transformation between data values and graphical element value. most often it is applied to position along x, y which is common, to log or sqrt, .. there are 3 numeric transformations.
Your turn
What are position adjustments? When would they be used?
positions shift the location some from original coordinates. most often used with bar charts to stack, or put side-by-side
Your turn
Use your cheat sheet to work out how to make a plot to explore the relationship between
Do you ever recline your seat when you fly? and Is it rude to recline your seat on a plane?
unique(fly_sub$`Is itrude to recline your seat on a plane?`)
# [1] "Yes, somewhat rude" "No, not rude at all" "Yes, very rude"
unique(fly_sub$`Do you ever recline your seat when you fly?`)
# [1] About half the time Usually Always
# [4] Once in a while Never
# Levels: Never Once in a while About half the time Usually Always
ggplot(fly_sub, aes(x=`Do you ever recline your seat when you fly?`)) +
geom_bar() +
facet_wrap(~`Is itrude to recline your seat on a plane?`, ncol=3) +
coord_flip()
ggplot(fly_sub, aes(x=`Do you ever recline your seat when you fly?`,
fill=`Is itrude to recline your seat on a plane?`)) +
geom_bar()
ggplot(fly_sub, aes(x=`Do you ever recline your seat when you fly?`,
fill=`Is itrude to recline your seat on a plane?`)) +
geom_bar(position="dodge")
Facets
ggplot(fly_sub,
aes(x=`In general, is itrude to bring a baby on a plane?`)) +
geom_bar() + coord_flip() + facet_wrap(~Gender)
Facets
fly_sub$Age <- factor(fly_sub$Age,
levels=c("18-29","30-44","45-60","> 60"))
ggplot(fly_sub,
aes(x=`In general, is itrude to bring a baby on a plane?`)) +
geom_bar() + coord_flip() + facet_grid(Age~Gender)
Color palettes - default
p <- ggplot(fly_sub,
aes(x=`In general, is itrude to bring a baby on a plane?`,
fill=Gender)) +
geom_bar(position="fill") + coord_flip() +
facet_wrap(~Age, ncol=5)
p
Color palettes - brewer
p + scale_fill_brewer(palette="Dark2") 
Color blind-proofing
What it looks like to a color-blind:
library(scales)
library(dichromat)
p1 <- p + theme(legend.position = "none")
clrs <- hue_pal()(3)
clrs <- dichromat(clrs)
p2 <- p + scale_fill_manual("", values=clrs) +
theme(legend.position = "none")
grid.arrange(p1, p2)
Perceptual principles
- Hierarchy of mappings: (first) position along an axis - (last) color (Cleveland, 1984; Heer and Bostock, 2009)
- Pre-attentive: Some elements are noticed before you even realise it.
- Color: (pre-attentive) palettes - qualitative, sequential, diverging.
- Proximity: Place elements for primary comparison close together.
- Change blindness: When focus is interrupted differences may not be noticed.
Hierarchy of mappings
- Position - common scale (BEST)
- Position - nonaligned scale
- Length, direction, angle
- Area
- Volume, curvature
- Shading, color (WORST)
Pre-attentive
Can you find the odd one out?
df <- data.frame(x=runif(100), y=runif(100),
cl=sample(c(rep("A", 1), rep("B", 99))))
ggplot(data=df, aes(x, y, shape=cl)) + theme_bw() +
geom_point() + theme(legend.position="None", aspect.ratio=1)
Is it easier now?
ggplot(data=df, aes(x, y, colour=cl)) +
geom_point() + theme_bw() +
theme(legend.position="None", aspect.ratio=1)
Color palettes
- Qualitative: categorical variables
- Sequential: low to high numeric values
- Diverging: negative to positive values
library(RColorBrewer)
display.brewer.all()
Proximity
ggplot(fly_sub, aes(x=`In general, is itrude to bring a baby on a plane?`,
fill=Gender)) +
geom_bar(position="fill") + coord_flip() + facet_wrap(~Age, ncol=5)
With this arrangement we can see proportion of gender within each rudeness category, and compare these across age groups. How could we arrange this differently?
Proximity
ggplot(fly_sub, aes(x=Gender,
fill=`In general, is itrude to bring a baby on a plane?`)) +
geom_bar(position="fill") + coord_flip() + facet_wrap(~Age, ncol=5) +
theme(legend.position="bottom")
ggplot(fly_sub, aes(x=Gender,
fill=`In general, is itrude to bring a baby on a plane?`)) +
geom_bar(position="fill") + coord_flip() + facet_wrap(~Age, ncol=5) + theme(legend.position="bottom")
What is different about the comparison now?
Another arrangement
ggplot(fly_sub, aes(x=Age,
fill=`In general, is itrude to bring a baby on a plane?`)) +
geom_bar(position="fill") + coord_flip() + facet_wrap(~Gender, ncol=5) +
theme(legend.position="bottom")
Themes
The ggthemes package has many different styles for the plots. Other packages such as xkcd, skittles, wes anderson, beyonce, ….
ggplot(fly_sub, aes(x=Gender,
fill=`In general, is itrude to bring a baby on a plane?`)) +
geom_bar(position="fill") + coord_flip() + facet_wrap(~Age, ncol=5) +
theme_xkcd() + theme(legend.position="bottom")
See the vignette for instructions on installing the xkcd font.
ggplot(fly_sub, aes(x=Gender,
fill=`In general, is itrude to bring a baby on a plane?`)) +
geom_bar(position="fill") + coord_flip() + facet_wrap(~Age, ncol=5) +
theme_xkcd() + theme(legend.position="bottom")